Tracking motion in 3D space

[Jassi]

I am stuck in my thesis work because I can't seem to come up with a way to track motion in 3D space AND stay consistent from day to day. The available technologies (IR Tracking system) give different results because even the slightest motion can change the output. This is a minor step but I have been banging my head against a wall for 2 weeks trying to find a solution.

Here's what I have :

1 Robotic arm
1 IR tracking camera (Polaris Accedo system by NDI) - Camera
Assorted accessories for the camera

My question is, can anyone think of a way to set up the hardware so that I can get readings that are consistent. I have to borrow the camera so the position changes after every use. Also, if there is a better system, please let me know.

 

[iotone]

aside from just buying a dedicated camera, maybe have something set up in your environment that can be used as calibration... an object of sorts that will allow you to be relatively consistent...

but for pure consistency, i think a dedicated system may be your best option

 

[robothouse77]

i work at a company that tracks a probe in 3D for medical purposes. They use IR, and have a new system which tracks using EM fields. It's much better than IR, as no line of sight is required. However, metallic interference can become a problem. Yet there are ways to overcome it.

 

[Matthias99]

Aw, man, I tried to do this for my thesis. IR cameras work pretty well -- if you can set them up permanently. If not, you'll have to deal with a little bit of variation each time. However, this can be dealt with -- for instance, when you first set up, you could put the robot arm in a fixed (known) position and take a snapshot of where the arm is in 'camera space'. Then, you can take all of the data from that day and translate/rotate it to correct for differences in camera position.

Does that make any sense? I can try to explain more if it didn't. It might also help if we had a few more details about what kind of experiments you're trying to do.

With my project, I had lots of restrictions -- we couldn't use magnetic sensors (because we had to use them in a constrained environment with cameras, sensors, etc. -- too much EMI). We couldn't use IR cameras, because the setup placed a sheet of plexiglass between the sensors and where we could put the cameras -- and then you get reflections that screw everything up. I ended up doing some work with a product called "ShapeTape" -- basically a flexible strip of bend/twist sensors that can be attached to an object and then used to track motion. It worked OK (but had major accuracy/consistency issues), but I ended up graduating before I got any real meaningful results (it was an optional honors thesis, so this wasn't the end of the world, and I still learned a *ton* and got two semesters of credit and a summer job out of it ). Lousy monkeys.

 

[Skyclad1uhm1]

I work for a company which mainly does 2D tracking on animals and humans, but also does 3D once in a while if the customer requires it. The easiest way might be to simply put a second camera in the system, so you can use the 2 2D images to create a 3D mapping, but you could also try placing the camera close enough for it to be able to notice the differences in size of the robot arm when it moves closer or further, and thus calculate the distance.
Or connect the camera to the arm, so you can simply let it measure how the camera itself moves by filming the surroundings.

 

[Jaimin]

I used to do research using a 3d motion camera. I don't know if the methods we used can apply to your project, but we also had to share the camera with the department. We would just always setup up a reference point, and all of our data sets were always the reference minus the collected data, so it didn't really matter where we started. This isn't the greatest for accuracy, but it was the best we could do, again, without having our own camera.

 

[Jassi]

Originally posted by: Jaimin
I used to do research using a 3d motion camera. I don't know if the methods we used can apply to your project, but we also had to share the camera with the department. We would just always setup up a reference point, and all of our data sets were always the reference minus the collected data, so it didn't really matter where we started. This isn't the greatest for accuracy, but it was the best we could do, again, without having our own camera.

Yea, I thought of it but its a *lot* of headache. Ah well, I guess its the only alternative

 

[harrkev]

Just a quick idea... Why not get a pair of cameras. If you use cheap $50 or so webcams, you can get color. Then, if you can control your environment, you can have your target object be one color that does not occur in the environment.

If you can get the data in your programming language of choice, you can convert RGB to HSV or something similar. Then target anything with a certain color, over a certain saturation and brightness. Take the centroid of the target object, and you are done. With two cameras, you can then triangulate (some basic trig is useful here).

 

[Jassi]

Hmmmm... do you have experience in this field?

Seems like a lot of programming in AR and I don't have the skill for it. I'll run it by my boss tho.

 

[mrscintilla]

I have a phd in the 2d tracking area. what harrkev talked about involving depth map and color conversation is considered the basic and the easy-to-do. Get some existing softwareware package (e.g., openCV), viola. But don't inspect any great accuracy though. If the environment does not change, tracking a only-moving object is not hard. IR cameras and some other hardware-assisted methods are way to go for 3D precision tracking. Any percisision stuff should not be handled by computer vision alone.

 

[seanp789]

Originally posted by: Jassi
I am stuck in my thesis work because I can't seem to come up with a way to track motion in 3D space AND stay consistent from day to day. The available technologies (IR Tracking system) give different results because even the slightest motion can change the output. This is a minor step but I have been banging my head against a wall for 2 weeks trying to find a solution.

Here's what I have :

1 Robotic arm
1 IR tracking camera (Polaris Accedo system by NDI) - Camera
Assorted accessories for the camera

My question is, can anyone think of a way to set up the hardware so that I can get readings that are consistent. I have to borrow the camera so the position changes after every use. Also, if there is a better system, please let me know.

My roommate is working on a project for NASA building a VR training system for MARS they use multiple cameras in different corners of the room to track 3D motion. A single camera can not effectively track i. They narrow down what they want to be tracked by placing neon colored pieces of tape(using a different color for each object) at the pivot and ends of object (like a skeleton in 3d animation.) The computer processes and tracks each point in 3d space. It has been a very effective. The system can track all motion but to make the distinction between different objects especially when they pass over or thru eachother you need the different colors for the computer to make the distinction.


Let me know if this helps or PM me if you want more detailed information/demo.

*EDIT* here is one of the actual write ups for the project with pictures.

http://www.cise.ufl.edu/research/vegroup/papers/i3d2005.pdf